Cross-Sibling Linkages in the Marriage Market

There is no field intervention in the conventional sense: this is a survey experiment in which the experimental variation is embedded in hypothetical vignettes administered within the interview. Each respondent evaluates a series of marriage-market scenarios in which we experimentally vary the circumstances of a hypothetical family, including the quality of a proposed match, the composition of the family's children, and the spacing between daughters. Respondents state beliefs about marriage-market outcomes and make advisory choices for these hypothetical families. Scenario content and presentation order are randomised across and within respondents. All scenarios concern fictional families.

The survey has three experimental modules, any two of which are administered to a respondent in one face-to-face sitting, together with a demographic questionnaire and a closing module on marriage norms. A verbal-consent question precedes all content; respondents may decline any individual question thereafter.

Respondents are screened into one of two belief modules: those with at least two daughters are eligible for the bride-side module, and those with at least one son are eligible for the groom-side module; where a household satisfies both screens, assignment follows the pre-printed respondent identifier.

In the bride-side module, respondents advise the parents of two unmarried daughters who receive a proposal for the elder daughter requiring a gift transfer exceeding half of the family's lifetime marriage savings. Respondents are presented with eight scenarios in which we vary: the proposed groom's quality (Low vs High), the presence of a brother (brother vs no brother), and the age gap between the daughters (2 vs 5 years), creating a 2 x 2 x 2 full factorial. In each scenario, respondents report: whether to accept the proposal; the expected distribution of the younger daughter's future match opportunities across three groom types had the elder daughter accepted; the effect of the elder's match on how groom-side families view the family; and the gift value the younger would require to marry a high-quality groom had the elder married the proposed groom. An enumerator-recorded protocol captures respondents who decline to name a gift value, distinguishing "no realistic amount", "she could not marry such a groom", comprehension difficulty, and refusal.

In the groom-side module, respondents advise the parents of a son (Medium or High type, varied between respondents) evaluating a proposal for the younger daughter of a family whose elder daughter has married, initially to an unknown groom. The module is blocked by family: for each of two hypothetical families (with vs without a son, order counterbalanced), respondents first answer baseline questions (acceptance likelihood on a 0-10 scale and a reservation gift value, at an offered gift of half the family's savings), then receive three sequential revelations of the elder daughter's realised match (High, Medium, or Low: Class 8 pass, factory labourer) with the offer adjusted to 0.3, 0.4, or 0.7 of savings respectively, restating both outcomes and additionally rating their view of the family (5-point scale) after each revelation. The respondent's own baseline answers for the current family are displayed for reference at each update. The three revelation orders within the first family follow a balanced set of all six permutations; the second family's revelations reverse the first family's order. The reservation-gift hesitation protocol distinguishes "no gift value would suffice", "gifts are irrelevant", comprehension difficulty, and refusal.

The allocation module is administered to all respondents. Respondents choose between an unequal split of the family's marriage savings favouring the elder daughter and an equal split, with both daughters' resulting match qualities fully specified on a 5-point labelled ladder with colour-coded scores. Cells vary whether the unequal split generates a cross-sibling gain, the age gap, and whether the family has a son (framed as easing the budget). Each respondent answers four of eight cells (2 x 2 x 2), drawn from sixteen pre-generated configurations balanced so that each cell is seen by an equal number of respondents; which side of the screen carries the equal split is randomised within respondent. Because outcomes are supplied, choices reveal allocation preferences uncontaminated by beliefs about the mapping from budgets to matches.

Between-respondent block factors: household lifetime marriage wealth W (Rs. 5,00,000 or 10,00,000); the unequal split (60-40 or 70-30), assigned jointly with W on the bride side and orthogonally to W on the groom side; and the advised son's quality (Medium vs High, groom side only). All assignments come from pre-generated lists balanced within sampling unit.

2026-07-19

2026-08-15

Bride-side module: (i) acceptance of the elder daughter's proposal; (ii) the believed distribution of the younger daughter's match opportunities across groom types; (iii) the perceived reputation effect of the elder's match; (iv) the believed gift value required for the younger daughter to marry a fixed high-quality groom.

Groom-side module: (v) acceptance likelihood (0-10); (vi) the reservation gift value at which the respondent would advise accepting the match; (vii) the updated assessment of the proposing family.

Allocation module: (viii) the binary choice between the unequal and equal budget split.

We are interested in three constructed objects. First, a respondent-level spillover belief index, constructed using within-respondent contrasts in the bride-side module. The change in the believed high-type share (outcome ii) and in the required gift for the younger daughter (outcome iv) as the elder daughter's proposed match varies across scenarios, holding all else fixed by construction. Second, a market-updating measure on the groom side. The within-respondent change in the reservation gift (outcome vi) between the baseline phase and each revealed elder-match state, interpretable as a revealed willingness to pay for family reputation; the belief-versus-market wedge is the difference between the bride-side dowry contrast and this groom-side contrast. Third, a belief-action pass-through coefficient. In the allocation module, the interaction between the respondent's spillover belief index and cell characteristics that reward tilting the budget toward the elder daughter; the shortfall of this interaction from full pass-through, in cells where tilting is stipulated to be advantageous, measures the extent to which fairness preferences over unequal giving block belief-consistent allocation.

The study is a within-subject vignette experiment conducted using a face-to-face survey of parents in two districts of Uttar Pradesh, sampled from 26 localities. Eligible respondents are assigned to one of two belief modules according to the composition of their own children, and all respondents complete a common budget-allocation module. Scenario attributes vary within respondent according to a full factorial design; scenario order is balanced across respondents using Williams Latin square designs, so that each scenario appears in each position, and follows every other scenario, with equal frequency. All experimental assignments are pre-generated by seeded computer routines, linked to anonymised respondent identifiers, and delivered automatically by the electronic survey instrument; neither enumerators nor respondents influence any assignment.

Assignment structure. Each sampled locality (PSU) contains 28 pre-assigned respondent identifiers: 1-14 for the bride side and 15-28 for the groom side. Within each side, ten identifiers per PSU constitute the core fieldwork sample and four are buffer, used only if a team over-recruits. Every identifier carries a pre-generated assignment row: a wealth-split block; a scenario-order row (one of eight bride-side orderings from a 4 x 4 Williams square over family types crossed with two groom sub-orders, or one of twelve groom-side orderings from the family counterbalance crossed with six revelation permutations); and four of the eight allocation cells with the randomised side labels. Assignments are balanced core-first: exact balance is enforced over the 260 core identifiers per side (blocks exactly equal; orderings, allocation configurations, and cell exposures equal to within one; scenario-by-position exposures equal to within one), with buffer identifiers balanced separately so that buffer usage cannot degrade core balance. Within each PSU, core identifiers span each block two or three times and no two identifiers share a full questionnaire profile. The electronic form retrieves all assignments by respondent identifier; identifiers failing lookup are rejected at entry.

Identification. Bride-side belief parameters are identified from within-respondent variation across the eight scenarios with respondent fixed effects; block-level magnitudes are absorbed and used for heterogeneity and scaling. Groom-side updating is identified from within-respondent changes between each family's baseline and its revelation states. The belief-market wedge subtracts the groom-side reservation-gift response to the elder's match from the bride-side believed dowry response, both defined at the fixed high-quality target. The allocation analysis interacts the respondent-level belief index with cell attributes; the cell in which the unequal option both raises total attainable quality and compresses the outcome gap separates preferences over unequal giving from preferences over unequal outcomes. Item non-response is permitted throughout and coded as refusal; hesitation-protocol reasons on gift-value questions are coded as informative censoring and analysed separately from refusal. Standard errors are clustered at the PSU level (26 clusters), with wild-cluster bootstrap as robustness; enumerator fixed effects are estimable because enumerators rotate across PSUs.

Piloting. Instruments were revised after a small pilot (rupee magnitudes, family-blocked presentation in both belief modules, allocation module reduced from twelve to eight cells with a 5-point ladder and randomised option sides, hesitation protocols, consent gate, universal item skip).

All experimental assignments were generated in office by seeded computer scripts and attached to pre-printed respondent identifiers before fieldwork. Within-respondent scenario orders follow Williams Latin square designs (bride side: a 4 x 4 square over family types crossed with counterbalanced groom sub-orders; groom side: counterbalanced family order crossed with a balanced set of revelation permutations). Between-respondent factors are assigned from pre-generated lists with exact balance enforced over the core sample within and across primary sampling units. PSU selection used systematic PPS with a seeded random start.

All experimental assignments are at the individual respondent level. Between-respondent factors (wealth-split block; groom-side son type; allocation split; allocation cell configuration) and within-respondent presentation orders (Williams Latin square rows; Phase-1 counterbalance) are assigned to anonymised respondent identifiers in pre-generated lists, stratified and balanced within primary sampling unit (locality). Sampling, as distinct from experimental assignment, is clustered: localities are selected by systematic PPS within district, and respondents are recruited within locality subject to screening; no treatment is assigned at the cluster level.

No

We survey in two districts of Uttar Pradesh (Lucknow and Prayagraj). Within each district we select 10 rural villages and 3 urban municipal wards by systematic probability-proportional-to-size sampling from the 2011 Census frame (minimum 500 households), reflecting the state's rural-urban population composition - 26 primary sampling units in total. In Lucknow district the rural frame excludes villages of Lucknow tehsil, which contains the municipal corporation. Many villages listed in the 2011 census have since become peri-urban. Two pre-specified frame-adjacent replacement PSUs are recorded for every selected PSU. We survey 20 respondents meeting screening criteria in each PSU.

Each Bride-Side respondent provides : 32 responses from the Belief Module and 6 responses from the Allocation Choice Module. Thus a total of 38 responses. Each Groom-Side respondent provides : 22 responses from the Belief Module and 6 responses from the Allocation Choice Module. Thus a total of 28 responses.

Approximately 500 total respondents: 250+ Bride-Side respondents and 250+ Groom-side respondents.